Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x1c349bd87b8>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x1c349c92be0>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.1.0
Default GPU Device: /gpu:0

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    input_ = tf.placeholder(dtype=tf.float32, shape=(None, image_width, image_height, image_channels), name="input")
    z_data = tf.placeholder(dtype=tf.float32, shape=(None, z_dim), name="z_data")
    lr = tf.placeholder(dtype=tf.float32, shape=(), name="learnings_rate")
    
    return input_, z_data, lr


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variabes in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).

def leaky_relu(x, alpha=0.1, name=None):
    return tf.maximum(x*alpha, x, name=name)
    
def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param image: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    keep_prob = 0.9
    
    with tf.variable_scope("discriminator", reuse=reuse):
        # images have a shape of 28x28xchannels
        conv = tf.layers.conv2d(images, filters=32, kernel_size=(7, 7), strides=(1, 1), padding="same", activation=None)
        drop = tf.nn.dropout(conv, keep_prob=keep_prob)
        relu = leaky_relu(drop)
        
        # 28x28x32
        conv = tf.layers.conv2d(relu, filters=64, kernel_size=(7, 7), strides=(2, 2), padding="same", activation=None)
        drop = tf.nn.dropout(conv, keep_prob=keep_prob)
        relu = leaky_relu(drop)
        
        # 14x14x64
        conv = tf.layers.conv2d(conv, filters=256, kernel_size=(7, 7), strides=(2, 2), padding="same", activation=None)
        drop = tf.nn.dropout(conv, keep_prob=keep_prob)
        relu = leaky_relu(drop)
        
        # 7x7x256
        flat = tf.reshape(relu, shape=(-1, 7*7*256))
        conn = tf.layers.dense(flat, 512)
        drop = tf.nn.dropout(conn, keep_prob=keep_prob)
        relu = leaky_relu(drop)
        
        # 512
        logit = tf.layers.dense(relu, 1)
        out = tf.sigmoid(logit)
        
    return out, logit


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variabes in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim, is_train=True):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    with tf.variable_scope("generator", reuse=not is_train):
        conn = tf.layers.dense(z, 512)
        relu = leaky_relu(conn)
        
        # 512
        conn = tf.layers.dense(relu, 7*7*512)
        conn_bn = tf.layers.batch_normalization(conn, training=is_train)
        conn_relu = leaky_relu(conn_bn)
        
        imagey = tf.reshape(conn_relu, shape=(-1, 7, 7, 512))
        
        # 7x7x512
        conv = tf.layers.conv2d_transpose(imagey, filters=256, kernel_size=(7, 7), strides=(2, 2), padding="same", activation=None)
        conv_bn = tf.layers.batch_normalization(conv, training=is_train)
        conv_relu = leaky_relu(conv_bn)
        
        # 14x14x256
        conv = tf.layers.conv2d_transpose(conv_relu, filters=128, kernel_size=(7, 7), strides=(2, 2), padding="same", activation=None)
        conv_bn = tf.layers.batch_normalization(conv, training=is_train)
        conv_relu = leaky_relu(conv_bn)
        
        # 28x28x128
        conv = tf.layers.conv2d_transpose(conv_relu, filters=64, kernel_size=(7, 7), strides=(1, 1), padding="same", activation=None)
        conv_bn = tf.layers.batch_normalization(conv, training=is_train)
        conv_relu = leaky_relu(conv_bn)
        
        # 28x28x64
        conv = tf.layers.conv2d_transpose(conv_relu, filters=out_channel_dim, kernel_size=(7, 7), strides=(1, 1), padding="same", activation=None)
        
        # 28x28xout_channel_dim
        out = tf.tanh(conv)
    
    return out


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    input_fake = generator(input_z, out_channel_dim=out_channel_dim, is_train=True)
    
    disc_real_out, disc_real_logits = discriminator(input_real, reuse=False)
    disc_fake_out, disc_fake_logits = discriminator(input_fake, reuse=True)
    
    loss = lambda **kwargs: tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(**kwargs))
    
    disc_real_loss = loss(logits=disc_real_logits, labels=tf.ones_like(disc_real_out) * tf.random_uniform(tf.shape(disc_real_out), 0.7, 1.2))
    disc_fake_loss = loss(logits=disc_fake_logits, labels=tf.zeros_like(disc_fake_out) * tf.random_uniform(tf.shape(disc_fake_out), 0.0, 0.3))
    disc_loss = disc_real_loss + disc_fake_loss
    
    gen_loss = loss(logits=disc_fake_logits, labels=tf.ones_like(disc_fake_out))
    
    return disc_loss, gen_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def minimize_with_clip(optimizer, loss, var_list):
    # Gradient Clipping
    gradients = optimizer.compute_gradients(loss, var_list=var_list)
    capped_gradients = [(tf.clip_by_value(grad, -1., 1.), var) for grad, var in gradients if grad is not None]
    train_op = optimizer.apply_gradients(capped_gradients)
    
    return train_op

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    trainable = tf.trainable_variables()
    disc_vars = [var for var in trainable if var.name.startswith('discriminator')]
    gen_vars = [var for var in trainable if var.name.startswith('generator')]
    
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        disc_optimizer = minimize_with_clip(
            tf.train.AdamOptimizer(learning_rate, beta1=beta1), loss=d_loss, var_list=disc_vars)
        gen_optimizer = minimize_with_clip(
            tf.train.AdamOptimizer(learning_rate, beta1=beta1), loss=g_loss, var_list=gen_vars)

    return disc_optimizer, gen_optimizer


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

# I'm going to pick the same random numbers every time, so hopefully we'll
# see a nice progression on digit/face quality for the same inputs. Generate
# a bit set of numbers to accomodate varying n_images and z_dim.
fixed_seed = np.random.uniform(-1, 1, size=[1024*4, 1024*4])


def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    
    # Pick the same random numbers every time.
    example_z = fixed_seed[:n_images, :z_dim]

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.figure(figsize=(8, 8))
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

run_i = 0

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    global run_i
    run_i += 1
    
    # Build model.
    nr_inputs, image_width, image_height, image_channels = data_shape

    input_, z_data, lr = model_inputs(image_width, image_height, image_channels, z_dim)
    discriminator_loss, generator_loss = model_loss(input_, z_data, image_channels)
    discriminator_optimizer, generator_optimizer = model_opt(discriminator_loss, generator_loss, learning_rate, beta1)
    
    def get_z():
        return np.random.uniform(-1, 1, size=(batch_size, z_dim))
    
    # Number of samples to show.
    nr_samples = 64

    # Train network.
    saver = tf.train.Saver()
    
    batch_i = 0
    
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            for batch_images in get_batches(batch_size):
                
                batch_i += 1
                
                # Images are in the range -0.5 to 0.5, bring them up to -1, 1.
                batch_images = batch_images * 2
                
                for i in range(4):
                    feed = {input_: batch_images, z_data: get_z(), lr: learning_rate}
                
                    sess.run([discriminator_optimizer, generator_optimizer], feed_dict=feed)

                if batch_i % 50 == 0:
                    discriminator_train_loss, generator_train_loss = sess.run([discriminator_loss, generator_loss], feed_dict=feed)
                    
                    print("Epoch {}, batch {}: discriminator loss {:.4f}, generator loss {:.4f}".format(
                        epoch_i, batch_i, discriminator_train_loss, generator_train_loss))
                    
                if batch_i % 500 == 0:
                    show_generator_output(sess, nr_samples, z_data, image_channels, data_image_mode)
                    
                    
        saver.save(sess, "./checkpoints/{}.ckpt".format(run_i))
        
        show_generator_output(sess, nr_samples, z_data, image_channels, data_image_mode)

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size = 8
z_dim = 128
learning_rate = 0.00005
beta1 = 0.4


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch 0, batch 50: discriminator loss 1.4702, generator loss 0.8260
Epoch 0, batch 100: discriminator loss 1.3579, generator loss 0.7304
Epoch 0, batch 150: discriminator loss 1.4894, generator loss 0.7484
Epoch 0, batch 200: discriminator loss 1.1997, generator loss 1.0099
Epoch 0, batch 250: discriminator loss 1.3366, generator loss 0.9190
Epoch 0, batch 300: discriminator loss 1.5027, generator loss 0.8448
Epoch 0, batch 350: discriminator loss 1.2923, generator loss 0.9495
Epoch 0, batch 400: discriminator loss 1.4012, generator loss 0.5639
Epoch 0, batch 450: discriminator loss 1.1645, generator loss 0.7681
Epoch 0, batch 500: discriminator loss 0.9175, generator loss 1.1246

Epoch 0, batch 550: discriminator loss 1.2021, generator loss 0.7888
Epoch 0, batch 600: discriminator loss 1.2492, generator loss 0.8226
Epoch 0, batch 650: discriminator loss 1.0140, generator loss 1.1256
Epoch 0, batch 700: discriminator loss 1.3785, generator loss 0.6191
Epoch 0, batch 750: discriminator loss 1.0294, generator loss 0.7388
Epoch 0, batch 800: discriminator loss 1.4409, generator loss 0.6583
Epoch 0, batch 850: discriminator loss 1.1206, generator loss 0.9054
Epoch 0, batch 900: discriminator loss 1.2938, generator loss 0.9354
Epoch 0, batch 950: discriminator loss 0.9649, generator loss 1.3275
Epoch 0, batch 1000: discriminator loss 1.2499, generator loss 0.8493

Epoch 0, batch 1050: discriminator loss 1.4521, generator loss 0.5868
Epoch 0, batch 1100: discriminator loss 1.0995, generator loss 1.1673
Epoch 0, batch 1150: discriminator loss 1.0918, generator loss 0.8069
Epoch 0, batch 1200: discriminator loss 0.9038, generator loss 0.8611
Epoch 0, batch 1250: discriminator loss 1.0685, generator loss 0.8431
Epoch 0, batch 1300: discriminator loss 1.1355, generator loss 0.7070
Epoch 0, batch 1350: discriminator loss 1.3905, generator loss 0.5239
Epoch 0, batch 1400: discriminator loss 1.1119, generator loss 1.0044
Epoch 0, batch 1450: discriminator loss 1.1361, generator loss 0.8889
Epoch 0, batch 1500: discriminator loss 1.2217, generator loss 0.8173

Epoch 0, batch 1550: discriminator loss 1.1757, generator loss 0.8104
Epoch 0, batch 1600: discriminator loss 1.1751, generator loss 0.6614
Epoch 0, batch 1650: discriminator loss 1.0838, generator loss 0.9045
Epoch 0, batch 1700: discriminator loss 1.1977, generator loss 0.7019
Epoch 0, batch 1750: discriminator loss 1.1552, generator loss 0.7322
Epoch 0, batch 1800: discriminator loss 1.0695, generator loss 0.8065
Epoch 0, batch 1850: discriminator loss 1.1928, generator loss 0.8155
Epoch 0, batch 1900: discriminator loss 1.1104, generator loss 0.9117
Epoch 0, batch 1950: discriminator loss 1.3608, generator loss 0.8063
Epoch 0, batch 2000: discriminator loss 1.2096, generator loss 0.9167

Epoch 0, batch 2050: discriminator loss 1.2769, generator loss 0.7535
Epoch 0, batch 2100: discriminator loss 1.3453, generator loss 0.7017
Epoch 0, batch 2150: discriminator loss 1.1356, generator loss 0.6943
Epoch 0, batch 2200: discriminator loss 1.0301, generator loss 0.8871
Epoch 0, batch 2250: discriminator loss 1.2633, generator loss 0.7594
Epoch 0, batch 2300: discriminator loss 0.9272, generator loss 0.9995
Epoch 0, batch 2350: discriminator loss 1.0897, generator loss 0.9012
Epoch 0, batch 2400: discriminator loss 0.8249, generator loss 0.8807
Epoch 0, batch 2450: discriminator loss 1.0726, generator loss 0.9132
Epoch 0, batch 2500: discriminator loss 1.1145, generator loss 0.7988

Epoch 0, batch 2550: discriminator loss 1.0640, generator loss 1.0159
Epoch 0, batch 2600: discriminator loss 1.2004, generator loss 0.8442
Epoch 0, batch 2650: discriminator loss 0.8754, generator loss 1.1567
Epoch 0, batch 2700: discriminator loss 1.0381, generator loss 1.1089
Epoch 0, batch 2750: discriminator loss 0.9795, generator loss 0.9436
Epoch 0, batch 2800: discriminator loss 1.1673, generator loss 0.7339
Epoch 0, batch 2850: discriminator loss 1.2471, generator loss 0.9151
Epoch 0, batch 2900: discriminator loss 1.2056, generator loss 0.7794
Epoch 0, batch 2950: discriminator loss 1.0304, generator loss 0.7953
Epoch 0, batch 3000: discriminator loss 0.9921, generator loss 1.0584

Epoch 0, batch 3050: discriminator loss 0.9995, generator loss 1.2215
Epoch 0, batch 3100: discriminator loss 1.0247, generator loss 1.0566
Epoch 0, batch 3150: discriminator loss 1.0612, generator loss 0.8665
Epoch 0, batch 3200: discriminator loss 0.8677, generator loss 1.5581
Epoch 0, batch 3250: discriminator loss 1.2394, generator loss 0.9427
Epoch 0, batch 3300: discriminator loss 1.2667, generator loss 0.8276
Epoch 0, batch 3350: discriminator loss 0.9263, generator loss 0.9748
Epoch 0, batch 3400: discriminator loss 0.8964, generator loss 1.1946
Epoch 0, batch 3450: discriminator loss 1.1783, generator loss 0.6758
Epoch 0, batch 3500: discriminator loss 1.1610, generator loss 0.7681

Epoch 0, batch 3550: discriminator loss 1.1113, generator loss 0.7973
Epoch 0, batch 3600: discriminator loss 1.0393, generator loss 0.8801
Epoch 0, batch 3650: discriminator loss 1.0135, generator loss 0.9639
Epoch 0, batch 3700: discriminator loss 1.0860, generator loss 0.9661
Epoch 0, batch 3750: discriminator loss 0.9085, generator loss 0.8731
Epoch 0, batch 3800: discriminator loss 0.8892, generator loss 0.9305
Epoch 0, batch 3850: discriminator loss 1.3330, generator loss 0.7178
Epoch 0, batch 3900: discriminator loss 0.9825, generator loss 1.1154
Epoch 0, batch 3950: discriminator loss 1.1494, generator loss 0.9210
Epoch 0, batch 4000: discriminator loss 1.0779, generator loss 0.8455

Epoch 0, batch 4050: discriminator loss 0.9012, generator loss 0.9366
Epoch 0, batch 4100: discriminator loss 1.0099, generator loss 0.7927
Epoch 0, batch 4150: discriminator loss 1.1881, generator loss 0.9961
Epoch 0, batch 4200: discriminator loss 1.4217, generator loss 0.6176
Epoch 0, batch 4250: discriminator loss 0.8579, generator loss 0.9430
Epoch 0, batch 4300: discriminator loss 0.9070, generator loss 1.0565
Epoch 0, batch 4350: discriminator loss 0.7467, generator loss 1.5268
Epoch 0, batch 4400: discriminator loss 0.9509, generator loss 1.0617
Epoch 0, batch 4450: discriminator loss 0.9191, generator loss 1.0440
Epoch 0, batch 4500: discriminator loss 1.1981, generator loss 1.1876

Epoch 0, batch 4550: discriminator loss 0.9127, generator loss 1.0674
Epoch 0, batch 4600: discriminator loss 0.9756, generator loss 0.9846
Epoch 0, batch 4650: discriminator loss 1.1409, generator loss 0.8915
Epoch 0, batch 4700: discriminator loss 0.7673, generator loss 1.2209
Epoch 0, batch 4750: discriminator loss 1.0405, generator loss 1.1713
Epoch 0, batch 4800: discriminator loss 0.6973, generator loss 1.2551
Epoch 0, batch 4850: discriminator loss 1.3926, generator loss 0.7116
Epoch 0, batch 4900: discriminator loss 0.8578, generator loss 1.1314
Epoch 0, batch 4950: discriminator loss 1.1336, generator loss 0.8282
Epoch 0, batch 5000: discriminator loss 1.1942, generator loss 0.8858

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 8
z_dim = 256
learning_rate = 0.00005
beta1 = 0.4


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 1

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Epoch 0, batch 50: discriminator loss 1.7307, generator loss 0.6713
Epoch 0, batch 100: discriminator loss 0.9678, generator loss 0.8291
Epoch 0, batch 150: discriminator loss 1.3861, generator loss 1.3628
Epoch 0, batch 200: discriminator loss 0.9368, generator loss 1.0031
Epoch 0, batch 250: discriminator loss 1.0190, generator loss 0.8442
Epoch 0, batch 300: discriminator loss 1.2343, generator loss 0.7130
Epoch 0, batch 350: discriminator loss 1.0499, generator loss 0.9125
Epoch 0, batch 400: discriminator loss 1.0744, generator loss 0.8880
Epoch 0, batch 450: discriminator loss 1.4319, generator loss 0.7379
Epoch 0, batch 500: discriminator loss 1.3440, generator loss 0.6649

Epoch 0, batch 550: discriminator loss 1.2442, generator loss 0.7469
Epoch 0, batch 600: discriminator loss 1.2853, generator loss 0.7051
Epoch 0, batch 650: discriminator loss 1.1562, generator loss 0.7578
Epoch 0, batch 700: discriminator loss 1.1113, generator loss 1.0420
Epoch 0, batch 750: discriminator loss 1.0920, generator loss 0.7247
Epoch 0, batch 800: discriminator loss 1.5871, generator loss 0.6565
Epoch 0, batch 850: discriminator loss 1.0127, generator loss 0.8711
Epoch 0, batch 900: discriminator loss 1.2410, generator loss 0.7222
Epoch 0, batch 950: discriminator loss 1.1866, generator loss 0.7165
Epoch 0, batch 1000: discriminator loss 1.1839, generator loss 0.7348

Epoch 0, batch 1050: discriminator loss 0.9942, generator loss 1.0198
Epoch 0, batch 1100: discriminator loss 0.9148, generator loss 0.9923
Epoch 0, batch 1150: discriminator loss 1.2019, generator loss 0.7627
Epoch 0, batch 1200: discriminator loss 0.6277, generator loss 1.1418
Epoch 0, batch 1250: discriminator loss 1.0106, generator loss 1.0278
Epoch 0, batch 1300: discriminator loss 1.1270, generator loss 0.8926
Epoch 0, batch 1350: discriminator loss 1.2245, generator loss 0.7910
Epoch 0, batch 1400: discriminator loss 1.3276, generator loss 0.7450
Epoch 0, batch 1450: discriminator loss 1.2741, generator loss 0.7097
Epoch 0, batch 1500: discriminator loss 1.2377, generator loss 0.6781

Epoch 0, batch 1550: discriminator loss 0.8856, generator loss 1.0068
Epoch 0, batch 1600: discriminator loss 1.1961, generator loss 0.7024
Epoch 0, batch 1650: discriminator loss 1.2664, generator loss 0.7019
Epoch 0, batch 1700: discriminator loss 0.9146, generator loss 0.9944
Epoch 0, batch 1750: discriminator loss 0.9912, generator loss 0.9294
Epoch 0, batch 1800: discriminator loss 1.2558, generator loss 0.6073
Epoch 0, batch 1850: discriminator loss 1.1483, generator loss 0.9872
Epoch 0, batch 1900: discriminator loss 1.1467, generator loss 0.7304
Epoch 0, batch 1950: discriminator loss 1.2169, generator loss 0.7152
Epoch 0, batch 2000: discriminator loss 1.1789, generator loss 0.7309

Epoch 0, batch 2050: discriminator loss 1.1660, generator loss 0.8837
Epoch 0, batch 2100: discriminator loss 1.1777, generator loss 1.0409
Epoch 0, batch 2150: discriminator loss 1.2984, generator loss 0.7701
Epoch 0, batch 2200: discriminator loss 1.1046, generator loss 0.8015
Epoch 0, batch 2250: discriminator loss 0.9597, generator loss 0.9181
Epoch 0, batch 2300: discriminator loss 0.9625, generator loss 0.9033
Epoch 0, batch 2350: discriminator loss 1.0929, generator loss 0.7470
Epoch 0, batch 2400: discriminator loss 1.0014, generator loss 0.7903
Epoch 0, batch 2450: discriminator loss 0.9458, generator loss 0.8800
Epoch 0, batch 2500: discriminator loss 1.2855, generator loss 0.7890

Epoch 0, batch 2550: discriminator loss 1.1051, generator loss 0.7942
Epoch 0, batch 2600: discriminator loss 1.3114, generator loss 0.6463
Epoch 0, batch 2650: discriminator loss 1.2920, generator loss 0.5705
Epoch 0, batch 2700: discriminator loss 0.9739, generator loss 0.8760
Epoch 0, batch 2750: discriminator loss 1.2751, generator loss 0.8023
Epoch 0, batch 2800: discriminator loss 1.2503, generator loss 0.6467
Epoch 0, batch 2850: discriminator loss 1.0456, generator loss 1.1138
Epoch 0, batch 2900: discriminator loss 1.1673, generator loss 0.7911
Epoch 0, batch 2950: discriminator loss 1.5143, generator loss 0.5771
Epoch 0, batch 3000: discriminator loss 1.2598, generator loss 0.7083

Epoch 0, batch 3050: discriminator loss 1.0968, generator loss 0.7534
Epoch 0, batch 3100: discriminator loss 0.9786, generator loss 0.7239
Epoch 0, batch 3150: discriminator loss 1.1635, generator loss 0.9081
Epoch 0, batch 3200: discriminator loss 1.2438, generator loss 0.9028
Epoch 0, batch 3250: discriminator loss 1.2264, generator loss 0.7687
Epoch 0, batch 3300: discriminator loss 1.3197, generator loss 0.7152
Epoch 0, batch 3350: discriminator loss 1.2498, generator loss 0.8146
Epoch 0, batch 3400: discriminator loss 1.1400, generator loss 0.9583
Epoch 0, batch 3450: discriminator loss 1.3911, generator loss 0.7142
Epoch 0, batch 3500: discriminator loss 1.0582, generator loss 0.8692

Epoch 0, batch 3550: discriminator loss 1.2254, generator loss 0.7467
Epoch 0, batch 3600: discriminator loss 0.9688, generator loss 0.9785
Epoch 0, batch 3650: discriminator loss 1.2910, generator loss 0.8373
Epoch 0, batch 3700: discriminator loss 1.1965, generator loss 0.7946
Epoch 0, batch 3750: discriminator loss 1.2109, generator loss 0.7928
Epoch 0, batch 3800: discriminator loss 1.0133, generator loss 0.7464
Epoch 0, batch 3850: discriminator loss 1.1469, generator loss 0.8603
Epoch 0, batch 3900: discriminator loss 1.3003, generator loss 0.7429
Epoch 0, batch 3950: discriminator loss 1.1279, generator loss 0.6005
Epoch 0, batch 4000: discriminator loss 1.0607, generator loss 0.8724

Epoch 0, batch 4050: discriminator loss 1.0987, generator loss 0.7609
Epoch 0, batch 4100: discriminator loss 1.2498, generator loss 0.8117
Epoch 0, batch 4150: discriminator loss 1.1921, generator loss 0.7510
Epoch 0, batch 4200: discriminator loss 1.0739, generator loss 0.8464
Epoch 0, batch 4250: discriminator loss 0.9840, generator loss 0.6884
Epoch 0, batch 4300: discriminator loss 1.2265, generator loss 0.8919
Epoch 0, batch 4350: discriminator loss 1.2553, generator loss 0.7879
Epoch 0, batch 4400: discriminator loss 1.1826, generator loss 0.7280
Epoch 0, batch 4450: discriminator loss 0.9965, generator loss 0.7844
Epoch 0, batch 4500: discriminator loss 1.2641, generator loss 0.6780

Epoch 0, batch 4550: discriminator loss 0.9166, generator loss 0.9921
Epoch 0, batch 4600: discriminator loss 1.1041, generator loss 0.6632
Epoch 0, batch 4650: discriminator loss 1.5199, generator loss 0.6178
Epoch 0, batch 4700: discriminator loss 1.3864, generator loss 0.7222
Epoch 0, batch 4750: discriminator loss 1.1273, generator loss 0.7610
Epoch 0, batch 4800: discriminator loss 1.1865, generator loss 0.7854
Epoch 0, batch 4850: discriminator loss 1.3045, generator loss 0.7518
Epoch 0, batch 4900: discriminator loss 1.3604, generator loss 0.8053
Epoch 0, batch 4950: discriminator loss 1.2258, generator loss 0.7149
Epoch 0, batch 5000: discriminator loss 1.0059, generator loss 0.7719

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.